shim: add support for containerd v2 metrics\nAdd support for v2 conta…

…inerd metrics in the shim, v2 metrics are only used when runsc is run with --system-cgroup=true.\nContainerd requires v2 metrics when the host is run with CGroupsV2.\nThis issue was noticed when attempting to gather metrics on AL2023 which defaults to CGroupsV2.\nFixes: google#11472

Signed-off-by: Champ-Goblem <[email protected]>

pkg/shim/runsc/BUILD

@@ -31,6 +31,7 @@ go_library(
         "@com_github_containerd_cgroups//:go_default_library",
         "@com_github_containerd_cgroups//stats/v1:go_default_library",
         "@com_github_containerd_cgroups//v2:go_default_library",
+        "@com_github_containerd_cgroups//v2/stats:go_default_library",
         "@com_github_containerd_console//:go_default_library",
         "@com_github_containerd_containerd//api/events:go_default_library",
         "@com_github_containerd_containerd//api/types/task:go_default_library",
@@ -47,6 +48,7 @@ go_library(
         "@com_github_containerd_errdefs//:go_default_library",
         "@com_github_containerd_fifo//:go_default_library",
         "@com_github_containerd_log//:go_default_library",
+        "@com_github_containerd_go_runc//:go_default_library",
         "@com_github_containerd_typeurl//:go_default_library",
         "@com_github_gogo_protobuf//types:go_default_library",
         "@com_github_opencontainers_runtime_spec//specs-go:go_default_library",

pkg/shim/runsc/service.go

@@ -18,6 +18,7 @@ package runsc
 import (
 	"context"
 	"fmt"
+	"github.com/containerd/go-runc"
 	"io"
 	"os"
 	"path/filepath"
@@ -29,6 +30,7 @@ import (
 	"github.com/containerd/cgroups"
 	cgroupsstats "github.com/containerd/cgroups/stats/v1"
 	cgroupsv2 "github.com/containerd/cgroups/v2"
+	cgroupsv2stats "github.com/containerd/cgroups/v2/stats"
 	"github.com/containerd/console"
 	"github.com/containerd/containerd/api/events"
 	"github.com/containerd/containerd/api/types/task"
@@ -49,7 +51,7 @@ import (
 	"github.com/sirupsen/logrus"
 	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/cleanup"
-	"gvisor.dev/gvisor/pkg/shim/runtimeoptions/v14"
+	v14 "gvisor.dev/gvisor/pkg/shim/runtimeoptions/v14"
 
 	"gvisor.dev/gvisor/pkg/shim/extension"
 	"gvisor.dev/gvisor/pkg/shim/proc"
@@ -660,6 +662,18 @@ func (s *runscService) Stats(ctx context.Context, r *taskAPI.StatsRequest) (*tas
 	// as runc.
 	//
 	// [0]: https://github.com/google/gvisor/blob/277a0d5a1fbe8272d4729c01ee4c6e374d047ebc/runsc/boot/events.go#L61-L81
+	return s.getStats(stats, r)
+}
+
+func (s *runscService) getStats(stats *runc.Stats, r *taskAPI.StatsRequest) (*taskAPI.StatsResponse, error) {
+	if s.opts.RunscConfig["systemd-cgroup"] == "true" {
+		return s.getV2Stats(stats, r)
+	} else {
+		return s.getV1Stats(stats, r)
+	}
+}
+
+func (s *runscService) getV1Stats(stats *runc.Stats, r *taskAPI.StatsRequest) (*taskAPI.StatsResponse, error) {
 	metrics := &cgroupsstats.Metrics{
 		CPU: &cgroupsstats.CPUStat{
 			Usage: &cgroupsstats.CPUUsage{
@@ -708,10 +722,45 @@ func (s *runscService) Stats(ctx context.Context, r *taskAPI.StatsRequest) (*tas
 	}
 	data, err := typeurl.MarshalAny(metrics)
 	if err != nil {
-		log.L.Debugf("Stats error, id: %s: %v", r.ID, err)
+		log.L.Debugf("Stats error v1, id: %s: %v", r.ID, err)
+		return nil, err
+	}
+	log.L.Debugf("Stats success v1, id: %s: %+v", r.ID, data)
+	return &taskAPI.StatsResponse{
+		Stats: data,
+	}, nil
+}
+
+func (s *runscService) getV2Stats(stats *runc.Stats, r *taskAPI.StatsRequest) (*taskAPI.StatsResponse, error) {
+	metrics := &cgroupsv2stats.Metrics{
+		// The CGroup V2 stats are in microseconds instead of nanoseconds so divide by 1000
+		CPU: &cgroupsv2stats.CPUStat{
+			UsageUsec:     stats.Cpu.Usage.Total / 1000,
+			UserUsec:      stats.Cpu.Usage.User / 1000,
+			SystemUsec:    stats.Cpu.Usage.Kernel / 1000,
+			NrPeriods:     stats.Cpu.Throttling.Periods,
+			NrThrottled:   stats.Cpu.Throttling.ThrottledPeriods,
+			ThrottledUsec: stats.Cpu.Throttling.ThrottledTime / 1000,
+		},
+		Memory: &cgroupsv2stats.MemoryStat{
+			Usage:      stats.Memory.Usage.Usage,
+			UsageLimit: stats.Memory.Usage.Limit,
+			SwapUsage:  stats.Memory.Swap.Usage,
+			SwapLimit:  stats.Memory.Swap.Limit,
+			Slab:       stats.Memory.Kernel.Usage,
+			File:       stats.Memory.Cache,
+		},
+		Pids: &cgroupsv2stats.PidsStat{
+			Current: stats.Pids.Current,
+			Limit:   stats.Pids.Limit,
+		},
+	}
+	data, err := typeurl.MarshalAny(metrics)
+	if err != nil {
+		log.L.Debugf("Stats error v2, id: %s: %v", r.ID, err)
 		return nil, err
 	}
-	log.L.Debugf("Stats success, id: %s: %+v", r.ID, data)
+	log.L.Debugf("Stats success v2, id: %s: %+v", r.ID, data)
 	return &taskAPI.StatsResponse{
 		Stats: data,
 	}, nil